4 ms·
Finding sources for input data is something I struggle with when building deep learning models. Out of curiosity, how did you go about programmatically accessi
by toomanyrichies 4y ago
Finding sources for input data is something I struggle with when building deep learning models. Out of curiosity, how did you go about programmatically accessing the music files for all 120M+ songs, in order to create your embedding vector? I can't imagine iTunes has an API which would let a person do that.
- pbronez 4y agoGood reminder of the value of Adversarial Interoperability https://www.eff.org/deeplinks/2019/10/adversarial-interoperability https://www.eff.org/deeplinks/2019/10/adversarial-interopera...
- scarface74 4y agoIf by”adversarial” you mean a publicly documented and freely available API that has been around in some form for two decades.
- proc0 4y agoAlso would like to know. I can't even listen to the full songs, and assuming I have to pay. I can't imagine buying 120 million songs, so it has to be some collab with iTunes.
- exikyut 4y agoThinking about both processing time and the difficulty of sustaining 120M downloads' worth of programmatic access, I wouldn't be surprised if this is actually trained on the track previews.
- yunwal 4y agoI’m almost positive it is. If you put in a song with a bunch of different styles, sections (e.g. bohemian rhapsody) the suggestions match the preview
- latexr 4y ago> so it has to be some collab with iTunes. There’s no way today’s Apple would allow such a collaboration. They’d just keep the feature and market it as part of Apple Music.
- lfkdev 4y agoProbably scraped them
- from 4y agoThey do, it’s just rate limited. See https://news.ycombinator.com/item?id=34641623 https://news.ycombinator.com/item?id=34641623